NaturalIntelligence/imglab

What it solves

ImgLab is a web-based tool designed to label images for object detection training. It simplifies the process of creating training data by allowing users to draw shapes and points on images to identify objects and landmarks.

How it works

Running directly in the browser with no prerequisites, the tool provides a canvas where users can annotate images using various shapes (circles, rectangles, polygons) and feature points. It supports exporting data in multiple formats compatible with popular ML frameworks, such as dlib XML, dlib pts, Pascal VOC, and COCO.

Who it’s for

It is intended for developers and researchers training object detectors, particularly those using dlib, who need a precise and lightweight way to label image datasets.

Highlights

• Multiple Format Support: Exports to dlib XML, dlib pts, Pascal VOC, and COCO.
• Auto-suggestion: Includes a feature to suggest labels to speed up the process.
• Flexible Annotations: Supports drawing feature points, circles, rectangles, and polygons, with the ability to drag or resize shapes.
• Browser-based: Platform independent and requires minimal CPU and memory.
• Landmark Management: Special tools for dlib users to adjust the order of landmarks and feature points.